Asrock Z370 Pro4

[mackerel]

Got this as the cheapest Z370 mobo that wasn't made by MSI. Assembled it with an i3 8350k, Noctua D15, 2x4GB Corsair something or other 2800/3000 depending on XMP setting. I managed to resurrect an existing Win7 install and away we go.

At stock, I found it curious that it was quite a bit slower than 6700k @ 4 GHz in Prime95 when either ram limited, or multi-thread. I'll have to revisit that.

On OC things started to look bad. There is no voltage reading from the board outside of bios. At least, none that hwinfo64 or Intel XTU can find. Haven't tried Asrock's own tool yet. You only get VID. Voltage setting can be done absolute in bios, or offset. Same in XTU. With tinkering so far, I've managed 5.1 GHz with some voltage setting I don't recall. http://hwbot.org/submission/3672256_

I haven't managed to go further than that yet, in part, I'm not confident what voltage is actually being applied to CPU. Thinking more, I might have an indirect way of checking that, but that'll have to wait until I have more time. The Noctua seems fine. Running stock P95 small FFT it peaked at 63C after 9 minutes, so I know I have headroom there.

Back to the mobo bios, I tried increasing current limit, and turning off power save features but they don't seem to have helped either.

Starting to wonder if I should have picked another mobo...

 

[Tír na nÓg]

I think software are not updated yet with all Z370 sensors.

 

[mackerel]

With up to +0.25v offset, I get reported VID almost up to 1.5V. Still not enough to run 5.2 at all. 5.1 highly unstable.

Now I'm focusing my efforts on stabilising 5.0. Cinebench R15 doesn't like +0.225v so far, with two soft crashes followed by a hard lockup. Maybe I'm too quick to criticise the mobo, and the simple reality might be that my sample of CPU just doesn't clock highly. My Skylake-X I've only managed to run at 4.8 to 4.9, so getting similar here. Was hoping 14++ would give an improvement over 14+. Maybe not at the limit.

 

[Johan45]

Put up a pic of the voltage section from the BIOS we'll see if someone id familiar with Asrock. I would try their software if nothing else seems to be working yet for monitoring. It might even work.
EDIT: try dropping the voltage from VID. When I left my x299 board on auto they really overvolted as I raised the clock. Based on kaby you should try around 1.3V for 5.0 and work from there. Intel can't handle excessive voltage very well

 

[mackerel]

I just managed to stabilise 5.0 for CB15, at least for long enough to complete one run. I had to run up voltage offset to +275mV. VID was crossing 1.5v at this point. Temps were still fine, peaking around 75C in CB15.

I'll try to get bios pics tomorrow I think. It's almost bed time and I want to get a CB15 submission into hwbot before that. Just turned up ram speed, now to cross my fingers for enough stability to do another run or few. The only successful run scored 819. Not a lot, but we only have 4c4t here. I'm hoping it'll be more interesting for single thread later...

 

[Tír na nÓg]

I've always been happy with Asrock mobo, eben the (what we call ) mid range.

I had maybe 6 or 7 of them, since Ivy bridge.

 

[mackerel]

Sticking to the +275mV offset, it seems stable with everything I've run so far...

Apart from the pure clock run, everything was run at 5 GHz, and I didn't seriously bother to optimise the results beond basic stuff like closing stuff and where possible turning up the priority. Just wanted to get a lot done... and still more to do.

BTW I'm not sure how long I'll stay there, as I'm leading by the simple fact it's the only submission. As soon as someone buys a better sample that might be stable at 5.1 instead of mine at 5.0, the only way is down. Or the sub ambient crowd will push me down fast regardless, if they're not too busy playing with 8700k's.

 

[mackerel]

Finally installed the Asrock OC/monitoring utility, and it does report Vcore. Now, how much do I believe it? At stock, idle, it was well below a volt, and under P95 small FFT, it went to around 1.15v. That seems in line with Skylake-X. If I put back on the +275mV offset, with just using CPU-Z stress for a quick bit of loading, it was reporting mostly 1.3x. Sounds about right? Also, it might mean I have a lot more voltage headroom than I thought I had!

The Asrock reported Vcore did not match anything in hwinfo64, CPU-Z or VID. Maybe some mobo reported value nothing else picks up (yet)?

 

[Tír na nÓg]

^Time will tell (or smoke )!

 

[trents]

Got this as the cheapest Z370 mobo that wasn't made by MSI. Assembled it with an i3 8350k, Noctua D15, 2x4GB Corsair something or other 2800/3000 depending on XMP setting. I managed to resurrect an existing Win7 install and away we go.

At stock, I found it curious that it was quite a bit slower than 6700k @ 4 GHz in Prime95 when either ram limited, or multi-thread. I'll have to revisit that.

On OC things started to look bad. There is no voltage reading from the board outside of bios. At least, none that hwinfo64 or Intel XTU can find. Haven't tried Asrock's own tool yet. You only get VID. Voltage setting can be done absolute in bios, or offset. Same in XTU. With tinkering so far, I've managed 5.1 GHz with some voltage setting I don't recall. http://hwbot.org/submission/3672256_

I haven't managed to go further than that yet, in part, I'm not confident what voltage is actually being applied to CPU. Thinking more, I might have an indirect way of checking that, but that'll have to wait until I have more time. The Noctua seems fine. Running stock P95 small FFT it peaked at 63C after 9 minutes, so I know I have headroom there.

Back to the mobo bios, I tried increasing current limit, and turning off power save features but they don't seem to have helped either.

Starting to wonder if I should have picked another mobo...

Not sure why you are surprised that an i3 would be slower than an i7? Do you mean per core?

 

[mackerel]

Not sure why you are surprised that an i3 would be slower than an i7? Do you mean per core?

Why would a 4 GHz quad core be slower than another 4 GHz quad core on the assumption both have same IPC? P95 has no benefit from HT so that is irrelevant. Cache size is same size on both. For small tasks, they were equal. For big tasks and/or multi-thread tasks, the i3 was a bit slower. Enough to say it wasn't just measurement variation, it was consistent across the range. This is suggestive of a difference in ram or cache performance. The ram in the 6700k was 2666 dual rank, the ram in the 8350k was 3000 single rank. To compare across them... I can't remember the exact rule of thumb conversion factor I had previously determined to allow for the rank, but I think the difference was too great in the 6700k's favour to be solely explained by ram.

The prime suspect therefore is the L3 clock. The i3 runs at 3700 stock, and I haven't changed that yet. A different 6700k in my main system is running 4000 stock, but the actual test system 6700k isn't online at moment so I can't double check it. That might be the difference explaining it. I wasn't going to write until I had tested it, but since you asked, above is the thinking.

Edit: I've retested the 8350k at 4000 cache, hardly changes results so the difference is still unexplained. Next step would be to swap ram to see if that is the factor.

 

[trents]

So the 8350k has four true cores? I haven't been keep up with Coffee Lake to be honest.

 

[mackerel]

i3-8350k is 4 GHz stock (no turbo) quad core, no HT, unlocked. Some have compared it against previous generation i5s. The 8350k specifically has 8MB of L3 cache more like previous i7's.

 

[mackerel]

After a delid with liquid metal, I got CPU temps down 12C on hottest core, around 8C on the others, at stock, ambient adjusted, P95 29.3 small FFT load. Difference should be even greater with heavier loads. That has given me the confidence to play a bit more with the OC.

At +275mV offset and 5.0 GHz, I've not had problems running anything other Prime95, which I had to drop to 4.8 to prevent errors. No crashes though. The Asrock utility reported around 1.40v there, compared to VID which was beyond 1.5. I don't know if it is the monitoring or the utility, but the voltage changes gradually when you apply load. I doubt it is responding that slowly, and it is more likely a measurement artefact so I'm ignoring that.

Taken first steps to trying 5.1 +325mV, which is the mobo sensor is correct would be around 1.45v. This is Cinebench R15 unstable, but it isn't far off stability. I got one run in, and a lockup 2nd run. Maybe a little more voltage will do it...

For a pure clock run, at +350mV offset I got a submission in at 5.2 GHz. It was unstable, actually crashed during submission process but it went through and I could recall it separately.

 

[EarthDog]

Any chance you can list what your end voltage is. Offset doesn't mean much when we don't know the starting voltage is (or have to find the needle in the haystack in the thread).

 

[Tír na nÓg]

You're getting lazy mate. Post #8

 

[EarthDog]

"Or have to find that needle in the haystack in the thread".

 

[Tír na nÓg]

"Or have to find that needle in the haystack in the thread".

Or I can find it for you!

 

[Halc]

According to Asrock website Z370 Pro4 does not have any USB 3.1 gen2 connectors, is this true? According to the manual it has a USB-C type connector, but the speed is supposedly USB 3.1 Gen1, not gen2.

 

[mackerel]

Only the Asrock utility seems to report a believable voltage. I don't leave it running all the time so quite simply I don't know the voltage unless I specifically go look. I can easily set the offset. Obviously the voltage will vary with load anyway.

Just for indication, at +275mV take it as near enough 1.40v as reported by the Asrock utility, or 1.5+ VID.